Build on Beam with your coding agent

One command to get started.

$ curl -fsSL https://www.beam.cloud/install | sh

Install or upgrade the CLI, sign in through your browser, and connect your workspace and coding agents. Add --vm to start a persistent VM with a desktop.

Read the agent briefing Agent directory

Deploy to Beam from the agent you already use

Setup configures the supported agents installed on your machine. To connect a remote MCP client, add the server URL and sign in through your browser. Choose your workspace and permissions; credentials refresh automatically.

Cursor

Cursor · IDE

Use Beam's MCP tools from Cursor's agent mode.

Agent mode with MCP tools
Background agents
Per-project or global MCP config

beam mcp install --client cursor

Claude Code

Anthropic · Terminal

Run the Beam CLI and use workspace MCP tools from Claude Code.

Runs beam deploy and beam logs directly
MCP tools for apps, logs, and metrics
Project-scoped .mcp.json config

beam mcp install --client claude-code

Codex

OpenAI · Terminal

OpenAI's terminal agent. Beam registers as an MCP server in ~/.codex/config.toml.

Command execution and file edits
MCP over Streamable HTTP
Runs the Beam CLI in its sandbox

beam mcp install --client codex

Windsurf

Codeium · IDE

Use Beam's MCP tools from Cascade in Windsurf.

Cascade agent with MCP tools
Multi-file edits
Global MCP config

beam mcp install --client windsurf

Claude Desktop

Anthropic · Desktop

Manage deployments, logs, and metrics through MCP over stdio.

Read logs, metrics, and request stats
Deploy and roll back from chat
Manage secrets and databases

beam mcp install --client claude-desktop

Any MCP client

Streamable HTTP · IDE, Terminal, Desktop

Add the server URL to an HTTP MCP client and connect your workspace through OAuth.

Stateless JSON-RPC, one POST per call
OAuth with automatic token refresh
Docs MCP needs no token

https://app.beam.cloud/api/v1/mcp?features=vm,core

Docs server, no token: https://docs.beam.cloud/mcp

Compute and storage for your agent

Sandboxes for untrusted code

Run code in an isolated environment that boots in seconds. Stop it when finished or save a snapshot.

Sub-second cold starts

Restore CPU and GPU containers from memory snapshots.

GPUs on demand

Add an RTX 4090 to any app with one argument, or rent a whole H100 machine. Pay only while it runs.

Persistent state

Durable disks, shared volumes, and managed Postgres, Redis, MySQL, and MongoDB.

MCP and CLI access

MCP tools for VMs, apps, logs, and metrics. Use the CLI to work with local files.

Any container image

A Dockerfile or a public image becomes a persistent service with a URL, disks, and secrets.

Logs and metrics

Read container logs, CPU and GPU metrics, and request stats through MCP.

Deployment history

Each deployment creates a version. Roll back or stop it with one MCP call.

Pay as you go

billed by the second

Give your agents their own machines

Start Building Read the docs